Robust Feature Matching and Selection Methods

1. INTRODUCTION Multisensor image registration is necessary in many applications of remote sensing imagery, whose crucial problem is how to establish the correspondence between the features extracted from the reference and input images. Most existing methods only consider how to extract features, however, the quality of the features are ignored. In this paper, we combine scale invariant feature transform (SIFT) and maximally stable extremal region (MSER) to initialize a coarse-to-fine matching to extract plenty of control points(CPs) pairs. A concept of distribution quality (DQ) is introduced to measure the distribution of CPs pairs, and experimental analysis is conducted to analyze the effects of CPs pairs number and DQ on the registration root mean square error(RMSE). An automatic feature matching and selection algorithm is then introduced, extensive experiments demonstrate the effectiveness of the proposed algorithm by aligning real multisensor images. 2. CONTROL PONTS QUALITY The registration error is determined by two factors: the number and the distribution of the CPs. When CPs number is fixed, the distribution of CPs plays key role in registration. Zhu [1] indicated the strong relations lies between the probability of correct matching and information entropy(IE). Here, we introduce the concept DQ by means of the IE of CPs. For quantitative analysis, we compute the spatial distribution quality center of the CPs and DQ is calculated as (1).